• No results found

The impact of the European cohesion policy on regional growth


Academic year: 2023

Share "The impact of the European cohesion policy on regional growth"


Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst


1 Research seminar: International Economics and Globalisation

Name: Jakub Konrad Kołek Student number: 13316729

The impact of the European cohesion policy on regional growth Supervisor: Dr. Péter Földvári

Word count: 13589


2 Statement of originality: This document is written by Student Jakub Kołek who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. UvA Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Table of contents

1.Introduction and research question ... 3

2. Characteristics of the cohesion policy ... 4

2.1 The different funds ... 4

2.2 Examples of projects ... 6

2.3 Distribution of funding across regions ... 7

3.Theoretical models of regional growth ... 9

3.1 The neoclassical model ... 9

3.2 Extensions to the Solow model ... 10

3.3 The main views regarding externalities... 10

3.4 New economic geography ... 11

3.5 Endogenous growth models and the role of knowledge ... 12

3.6 Heterogenous individuals and firms ... 12

3.7 The impact of regional policy in a heterogenous economy ... 13

4. Existing empirical studies ... 14

4.1 Empirical evaluations of the cohesion funds... 14

4.2 The factors impacting the efficiency of grants ... 15

4.3 The ecological aspect of the cohesion subsidies ... 16

4.4 The implications of the empirical assessments ... 16

5. Data ... 18

5.1 Sample ... 18

6. The empirical exercise ... 20

6.1 Regression specification ... 20

6.2 Estimation approach ... 21

6.3 Estimating the long-term effects ... 22

6.4 Results ... 24

7. Conclusions ... 28

7.1 Policy implications ... 29

7.2 Avenues for further research ... 30

Bibliography ... 31



1.Introduction and research question

There are stark differences in levels of wealth and development between the regions of the European Union. There are two main factors that shape the dynamics of these discrepancies, each of them acting in opposing directions. First of all, according to the Neoclassical Growth Theory, economic integration makes these differences even out over time due to the mobility of production factors, leading to convergence in output and employment.

Subsequently, a countervailing force is described in the New Economic Geography literature, which explains that certain localities benefit from strong positive externalities and internal economies of scale, that result in the concentration of economic activity in comparatively larger agglomerations. Both of the theories have distinct merits in explaining the situation.

However, the EU still experiences severe disparities between regions, the richest areas have a GDP per capita up to four times as high as the most deprived1. Although the Neoclassical assumptions that the convergence process between regions should gradually occur are partially supported by the empirical data (Goecke and Hüther, 2016), the majority of the disparities still remain, even between the founding members of the Union.

The European Union has mounted a policy response with a purpose of mitigating these disparities. It is a set of funding measures under the general name of the “Cohesion Policy”

which fits into the European Structural Funds framework. In the 2014-2020 period, up to 34% of the EU budget was allocated to these programmes (Darvas et al, 2019). Such a large expenditure raises the need for its performance review, which should be conducted both by the specialised European and national institutions and by independent researchers. It also provides an opportunity to study the impact of a large public subsidies programme on the economic systems of European Regions.

This thesis conducts an econometric evaluation of the effects of EU regional cohesion funds expenditure on the economic growth of the European NUTS 2 regions in terms of the gross domestic product per worker (PPP) between 2004 and 2018. The analysis takes into account both the immediate effects as well as those which arise in longer timeframes, examining them using a distributed lag model. The efficiencies of the subsidies are also evaluated in function on the specific funding programme through which they was accorded. The study examines the period from the enlargement of the European Union by ten new member- states in 2004 until 2018. The end date is limited by the current availability of the data on cohesion expenditure.

The regression specification is based on the neoclassical growth framework. This setup was first developed by Solow (1956) and augmented by Mankiw, Romer and Weil (1992) to include the accumulation of human capital. Islam (1995) was the first to apply it to a panel data format. The model includes a lagged component of the dependent variable which leads to endogeneity issues. Therefore, dynamic panel estimation techniques founded on the instrumental variables approach are used to mitigate the bias that would arise in an ordinary least squares regression (Nickell, 1981).

1“Regional statistics by NUTS classification” database, Eurostat: https://ec.europa.eu/eurostat/web/regions/data/database



2. Characteristics of the cohesion policy

The European Union distributes grants and loans to underdeveloped regions with the main purpose of mitigating the disparities between them. The financial support is allocated through various programmes which follow a general strategy under the name “Cohesion Policy”. Since its beginning in 1989, this policy consisted of various funds and programmes with multiple modifications in rules and nomenclatures, however, its fundamental principles remain constant. The funding aims to improve the economic status of the poorer regions by investing in their competitiveness, funding research and development activities as well as developing the infrastructure and public utilities. Other goals include supporting environmental protection and the development of tourism. Grants require co-financing and can be received by local governments, non-government organisations or private companies.

The expenditures are scheduled in 7-year programming cycles with the adjustment of rules happening in between. This thesis examines the 2004-2018 period which spans across parts of three different cycles, 2000-2006, 2007-2013 and 2014-2020.

After the 2008 recession and the protracted period of sluggish recovery afterwards, the regional subsidies assumed a supplementary objective, not only to even out the differences between regions but also to try to stimulate economic growth within the European Union.

This was visible in an increased amount of investments aiming to induce growth during the 2014-2020 programming period. Moreover, due to increasing concerns about the climate change, a large proportion of the funds was allocated towards decarbonisation and environmental protection (European Commission, 2015).

2.1 The different funds

This thesis examines the effects of the cohesion policy funds on the economic growth of European regions during the 2004-2018 period. During this time, the cohesion subsidies were administered through three main programmes (European Commission, 2015): the European Regional Development Fund (ERDF), the Cohesion Fund (CF) and the European Social Fund (ESF). A supplementary programme, the Youth Employment Initiative (YEI) was launched in 2012, however, due to its short time of operation and relatively low extent, it had too little of an impact to be examined in this thesis. These cohesion policy funds formed the main part of the European Structural and Investment Funds (ESIF) framework, which included additionally the agricultural and maritime policy funds as well as a fund for the most deprived members of the society.

The particular funds diverge by their competences and purposes, the European Regional Development Fund and the Cohesion Funds concentrates mainly on investments in the public and private sectors in order to improve the infrastructure, support development and protect the natural environment. The European Social Fund focusses on providing education, training and supporting the creation of jobs, with the purpose of improving the social cohesion of the Union. The Youth Employment Initiative provides support to regions with high rates of unemployment amongst young people aged from 18 to 24. During the studied period, the total financial commitments of the EU on the cohesion policy amounted to 364 billion euro out of which 55% was promised through the ERDF, 23% through the ESF, 20%

through the CF and 1% through the YEI (Darvas et al, 2019). Almost all of the funding


5 requires co-financing from the local entities. The policy is categorised across the NUTS 2 regions, which are the basic statistical areas defined by the Eurostat and used for EU policy purposes. They are usually formed from a group of national administrative units and their populations range from 800 thousands to 3 million.

Although most of the funds operate in all of the European regions, their financing rules are set in a way which supports mainly the poorer regions thus realising their main goal which is mitigating the disparities between regions. The NUTS2 regions are classified in three distinct categories, the More Developed, Less Developed and Transition regions, each of the groups can receive different amounts of support from the ERDF and the ESF. The allocation happens at the beginning of each programming period. For example, the eligibility criteria for the 2014-2020 period, which are based on 2014 GDP figures are visible on the map below:

Figure 1: Categories of regions for the ERDF and ESF in the 2014-2020 period

Created with mapchart.net, source: European Commission, 2015. “European Structural and Investment Funds 2014-2020: official texts and commentaries”. Directorate General of Regional and Urban Policy

The ERDF is distributing the largest part of its funds to the Less Developed regions, which have below 75% of the average European GDP per capita. The policy was formerly called

“objective 1 funding” in the 2000-2006 period and “convergence objective” in the later periods. A smaller extent of support was provided to the Transition regions which range from 75% to 90% of the EU average.


6 The European Social Fund offers a larger share of co-funding to projects in the regions with lower levels of per capita output. The funding rates vary from 50% up to 85%, and in some special cases, go up as high as 95% (European Commission, 2015). Hence, poorer regions have to pay a lower own contribution in order to conduct a project.

The Cohesion Fund acts in a different way, targeting the disparities at a country level. It issues grants in countries that have below 90% of the average European per capita output, however, the funds are still administered at the NUTS 2 level. A combination of the funding rules results in allocating a large expenditure through the three programmes to the comparatively less developed areas, aiming to reinforce the cohesion between the different parts of the EU.

2.2 Examples of projects

It is helpful to look at the specific types of projects conducted by the different funds in order to visualise what they consist of and how they can influence economic growth. The table presented below describes the main objectives of the particular programmes.

Table 1: Main activities realised by funds


Transport infrastructure: railways, airports, roads, ports Water and environment

Energy production and distribution

Promotion of tourism, culture and arts funding

Healthcare: new equipment and specialised treatments Education and scientific research

Research and development activities (R&D) General business support

Reduction of carbon emissions


Transport infrastructure: railways, airports, roads, ports Trans-European transport networks

Public utilities: refuse and waste treatment

Environment: reforestation, erosion control, nature conservation Water: dams, irrigation, aqueducts etc.

Reduction of carbon emissions


Support for the disabled: mobility, accessibility, hearing aids

Labour market trainings: soft-skills, technology, specialised qualifications Vocational formation for adults

Professional training and support for prisoners

Retaining children and adolescents in schools, preventing drop-outs Promoting entrepreneurship

Supporting social enterprises and farms

Improving the efficiency of regional institutions

Providing new tools for spatial planners, social workers etc.

Source: own selection based on European Commission (2015) and the “Major projects in the framework of the Structural Funds and the Cohesion Fund” policy document; Due to the large amount and high variety of projects, this list serves only as a general indication and does not describe all of the purposes of the Cohesion Policy expenditure.


7 The European Regional Development Fund is the most versatile as it is financing a wide range of activities involving infrastructure development, regional promotion, R&D and business support. The funding is offered for a broad range of private and public entities which are co-managing the projects. The Cohesion Fund partially overlaps with the activities of the ERDF, however, it is more specialised in improving the transport infrastructure and in the environmental protection. The European Social Fund is specialized in the labour market activities such as the professional trainings and the acquisition of skills, with a special focus on the socially excluded groups such as people with disabilities. The projects target the vulnerable people and try to integrate them best to the society. In the current 2021-2027 programming period, the programme is joined with the Fund of European Aid for the Most Deprived, which was earlier an independent programme within the Structural Funds.

Together, they form now the “European Social Fund plus” which expands the ESF activities to the persons which require aid for basic vital needs, such as the homeless.

The nature of the projects performed by the different funds vary greatly, the same applies for their differing budgets as well as the regions they are targeting. This is why, they are likely to have different consequences for the economic growth. This divergence will be closer examined in the regression analysis presented in the section 6.

2.3 Distribution of funding across regions

Figure 2: Distribution of Cohesion Policy expenditure per capita in the EU regions

Source: “Historic EU payments - regionalised and modelled” dataset, European Commission, may 2020

The bulk of the regions received low amounts of funding, a little more than a half received less than 1000 euro per inhabitant over a 14 years period. The funds are then gradually phasing out, with gradually smaller numbers of regions receiving higher payments. There are some visible outliers above the 6000 EUR value, however, they do not change the bigger picture, the majority of the EU regions are receiving a moderate amount of support.


8 The disparities in the amount of the funding are a consequence of several factors. The principal factor is the main goal of the cohesion policy which offers additional funds to the regions and countries which are less-developed in terms of GDP per capita. As discussed in section 2.1, the countries which have below 90% of average GDP per capita can benefit of the Cohesion Fund and the regions which are below 75% of the average GDP per capita can obtain a much larger funding from the European Regional Development Fund only operate in selected regions of the European Union. Furthermore, the cohesion funding programmes offer different co-financing rates depending on the income per capita criterium. Therefore, the regions with lower GDP per capita values will receive more subsidies for the same amount of projects compared to their richer counterpart.

Finally, the eligibility is a necessary but not a sufficient condition to receive funds. The regional institutions and private sector entities have to actively apply and propose projects in order to access the resources. This is why some regions are able to receive much higher shares of investments. It is worth bearing in mind that the financial commitments obtained by regions do not have to be always fully used because a part of the projects might stop before or during their realisation periods. A large part of the projects can also decide to spend allocated funds only after their original programming period. According to the

“Cohesion Policy financial implementation timeseries”, only 52% of the accorded funds for the 2014-2020 programming cycle were spent by the end of 20202.

2 Source: European Commission, Open Data Portal for the European Structural Investment Funds https://cohesiondata.ec.europa.eu/2014-2020_cohesion_overview#



3.Theoretical models of regional growth

This section will introduce the most relevant theoretical models of regional economic growth and find what insights they can give about the cohesion policy investments. It is especially important to assess the implications they provide about for the divergence of regions and whether the structural investments are a suitable instrument to tackle this inequality. Furthermore, with relation to the main research question, this literature review will assess what are the conclusions that the different theories reveal about the impact of the regional funds on economic growth.

3.1 The neoclassical model

Most models of regional development originate from broader theories of economic growth such as the framework developed by Solow (1956), which is likely the most influential contribution to the field. In its fundamental version, the model assumes a Cobb-Douglass production function based on two production factors with constant returns to scale.

𝑌 = 𝐴𝐹(𝐾; 𝐿)

There are two production factors, labour 𝐿 and capital 𝐾, both of them exhibit decreasing returns to scale and the total factor productivity level 𝐴 is exogenous. This parameter reflects the efficiency of production resulting from the technology, work organisation and the social institutions employed. A proportion of the output 𝑠𝑌 where 𝑠𝜖(0; 1) is saved and invested in each period and the rest is consumed. The investments fuel the capital stock which depreciates at a fixed rate δ and population grows at the fixed rate 𝑛. The economic growth is measured in terms of the increase of the per capita output, denoted by 𝑦 =𝑌

𝐿. Positive growth can occur long as the effect of the capital accumulation is larger than the impacts of the population growth and the depreciation. This values will eventually equalize at the output level 𝑌 because of the diminishing marginal returns to capital. When this happens, the economy reaches a steady state equilibrium in which the economic growth can only occur at the pace of the technological progress. It is worth remarking, that economies do not have to move towards absolute convergence. Depending on their parameters, they will have different natural steady states. Ceteris paribus, a greater savings rate leads to a higher stationary state income (although not necessarily higher consumption), but a higher demographic increase results in lower per capita output.

In this setting, the European regional investments cannot change the long-term economic growth, they can only speed-up the convergence to the steady state and potentially impact the steady state level. Dall’Erba and Le Gallo (2008) argue that within a common market, allocating additional funding to lower-income regions can only increase temporarily increase the rate of convergence which happens thanks to the mobility of capital and labour. This is due to the fact that these less-developed areas should have a higher marginal product and thus offer a higher return on investment. Moreover, the workers will migrate to richer areas, driving their wages down while the resulting lack of labour will increase them in the poorer localities. These processes should naturally even out the differences. However, the structural funds could still have an effect on the steady states of regions if they would be able to increase the investment rates by themselves or by crowding-in more private-sector funds.



3.2 Extensions to the Solow model

Despite its relative simplicity, the neoclassical model has a good fit to the empirical data and can explain most of the variance in per capita incomes (Bernanke 2002). The basic framework has been modified multiple times to increase its complexity and to achieve a better representation of reality. Most notably, Mankiw, Romer and Weil (1992) introduced the human capital level 𝐻 to the production function:

𝑌 = 𝐴𝐹(𝐾; 𝐿; 𝐻)

In this case, the adjustment process includes additionally the accumulation of human capital through education and the acquisition of skills. The structural funds, especially the European Social Funds can provide additional educational opportunities and professional formation which, in turn will increase the steady state income of the targeted regions.

More recent generations of neoclassical models recognize the existence of the other factors which can shape the long-term per capita output of a particular regions. These studies consider a notion of “convergence clubs” which are the groupings of regions with similar spatial characteristics (Corrado, Martin and Weeks 2005). The underlying assumption is that some localities which have better access to resources and better locations for trade will have a higher growth potential than the peripheries. Therefore, output levels will tend to converge to the values proper for their groups.

3.3 The main views regarding externalities

Despite its compelling advantages, the basic neoclassical model does not explain why some regions would have a higher concentration of economic activity and why agglomerations are formed. In order to find the answer to this question, it is necessary to outline the theories which explain the benefits related to collocation of production and distribution of goods and service. Economic activities tend to concentrate in places where they can benefit of the positive externalities which arise from the collocations between producers or consumers.

The theories originating in the works of Marshall (1870), Arrow (1970) and Romer (1986), suggest that these spillovers effects occur mainly when areas specialise in the production of a particular good of service. Therefore, the producers can benefit of positive external economies due to a better access to a common pool of specialized workers and spillovers of technical knowledge within the region. Furthermore, clustering in large centres helps to access better-developed transport network and spread the large fixed costs of their maintenance. All these factors result in a decrease of transaction and distribution costs which make the enterprises operating within these areas more profitable, which in turns attracts more firms and workers. This tendency is counterbalanced by the high costs of operating in such locations. The described specialisation economies of scale can explain how industry clusters work, but they are unable to explain why they form. It is unclear whether they appear thanks to the positive externalities or are just reinforced by them.


11 Jacobs (1970) proposes a different approach in which the positive externalities do not come from specialisation but, on the contrary, from the diversification of activities. A high level of heterogeneity between sectors can be another potential source of positive externalities. A wider array of economic activities can create synergies between different sectors and industries, especially when technologies and organisational structures originating from one sector can be applied in others. Cooperation between various branches of industry can also result in more product innovations and new start-up companies (Porter, 1998). Moreover, cities offer a network of business services, research and development facilities, educational institutions and cultural establishments which can improve the efficiency and the creativity of workers.

3.4 New economic geography

Unlike previously discussed theories that focused on external economies of scale as reasons for the concentration of economic activity, the New Economic Geography approach focuses on the reasons which occur within individual producers. This diverging line of thinking appeared in the 1990’s with the works of Paul Krugman and uses the imperfect competition model as its core framework. Each firm produces an individual and differentiated product and the consumers can substitute between them to a certain degree. The companies can then profit from internal economies of scale without creating monopolies, however, this process favours the creation of a smaller number of larger entities than within a perfect competition (Harris, 2011).

In the version of the model which focusses on labour mobility, regions with higher amount of economic activity enjoy the internal economies of scale which increases their demand for labour. Workers migrate from other regions and in turn stimulate the home market even more. Krugman (1998) presents a highly stylized model which results in extreme outcomes where all of production can be concentrated in one locality or dispersed depending on the transport costs. If these are low enough, one of the regions can profit from the internal economies of scale and concentrate in itself the production for both regions. The model is relatively simple and only applies to a partial equilibrium setup with no money or capital markets, but it can serve as a starting point for the discussion on the existence of multiple possible equilibria. The New Economic Geography entails that is not always predetermined which cities will become the most important urban centres. Therefore, adequate policy can change the outcomes for given regions, which is also the rationale for the European cohesion policy.

The efficiency of the funding depends on whether it will be able to promote growth in the peripheral regions and counteract the processes which lead to the displacement of economic activity by enabling the poorer regions to benefit from internal economies of scale. The main difficulties related to this approach are that it is very difficult to test empirically and to provide reliable answers about regional growth. However, works such as Breinlich (2006) find some spatial structures within the EU are consistent with the model.

The concerns about the distribution of economic activity across regions are valid and should be taken into account in the policies and in the empirical studies.



3.5 Endogenous growth models and the role of knowledge

The endogenous economic growth theories were first introduced by Romer (1990). This approach differs from the Solow model in that technology growth is no longer exogenous, instead, it depends on the amount of R&D activities. Human capital is allocated to research when the sum of discounted returns from expected innovations outweigh the alternative cost of the discounted wage payments for the workers. Contrarily to labour and capital, knowledge does not exhibit diminishing marginal returns, which allows for sustainable economic growth which is fully generated in the model. When it comes to the implications for the impacts of cohesion policy investments, this theory suggests that the share of the funding which is allocated to the R&D objectives should be especially important for regional growth, as well as the economic development of the entire European Union. Furthermore, the initiatives which increase the education and skills level such as many of the ESF projects should have higher positive effects for growth, as the increase in human capital can lead to a higher R&D level.

The Innovation systems literature looks at these issues from a broader perspective.

According to the theories, the abilities of regions to innovate and to learn new technologies depend not only on the economic and technological factors but also on the institutional environment and the social systems (Cooke 1997). All these dimensions are interconnected and changes in each of them are also affecting others. The main takeaways from these frameworks are that the relations between the public and private sectors, as well as education and research matter for economic growth. Particularly, the quality of regional institutions and cooperation between sectors are vital for achieving development. This is further confirmed by empirical studies such as Rodrigues-Posse and Garzillazo (2015), which finds that regions with higher institutional quality indicators obtain higher benefits from the EU regional funds.

3.6 Heterogenous individuals and firms

As the economy becomes increasingly knowledge-driven, the reasons agglomerations arise are not limited to the material aspects but also to the spread of knowledge. The spread of information is the most efficient in close proximity due to the tacit character of ideas and skills. Recent developments take into account not only this aspect but also the heterogeneity of individuals and firms, which can differ in terms of their skills and learning capabilities.

Davis and Dingel (2019) propose a theoretical model in which agglomerations are created on the basis of the exchanges of ideas between firms. Skilled workers producing tradeable goods create positive spillovers in form of new ideas which are improving the efficiency of other skilled labour in a particular city. The positive externalities in turn attract more workers with a higher skill level. In consequence more skilled workers move to comparatively larger cities. The same authors build on their basic framework in Davis and Dingel (2020). This extended general equilibrium model allows for heterogeneities not only between cities but also between areas in a particular municipality. More populous cities have a higher employment in skill-intensive sectors. These sectors create positive externalities which makes it more attractive to move to this city. This creates a feedback effect, as more productive workers move to larger agglomerations and further increase the


13 size of the positive spillovers. However, relocating to these highly-sought locations is costly, which limits the migrations. The estimates of the model are then verified with the empirical data from 270 American metropolitan areas. The results are consistent with the statistics in terms of the distribution of education attainment levels in the population.

Gaubert (2018) presents a theoretical model of an economy with heterogenous firms, which differ in terms of total factor productivity, in order to find the patterns according to which they are distributed across cities. The model assumes that the differentials in total factor productivity between localities originate from two different effects. Firstly, businesses in large agglomerations benefit from positive externalities related to the presence of other firms, which allow them to be more efficient. Secondly, more productive companies will tend to move to larger cities than in order to profit from the spillovers. The presence of more productive firms in larger cities increases the externalities even more. Similarly to Davis and Dingel (2020), this circular mechanism is counteracted by the increasing costs of relocating to the most desirable locations. The author then conducts a simulation of the general equilibrium in order to disentangle the magnitude of the two effects. The results indicate that both of them are similarly strong, each of them accounting by itself for about half of the observed total factor productivity differentials.

3.7 The impact of regional policy in a heterogenous economy

Gaubert (2018) applies its model in order to estimate the impact of place-based subsidies using data for the French “ZFU” program, which was targeting businesses in low-income urban areas. The participants received a 12% subsidy to their profits financed from a tax on other enterprises. In the simulation, the funding resulted in a 19% increase in the number of businesses in these areas which is consistent with the empirical results (Mayer et al. 2015).

Nevertheless, the overall welfare effects are negative because the majority of these firms did not appear spontaneously but moved from other localities and became less productive after moving. The project aimed to reduce the inequality between areas. Unexpectedly, the subsidies increased the regional income inequality measured by the GINI coefficient. The reason for this increase, is that the firms which relocated to the subsidised areas were originating from medium-sized cities with average incomes. Their departure harmed these areas and widened the gap between them and the metropoles.

These results pose the question whether the medium-income regions in the EU are not disadvantaged due to the cohesion funds being allocated mostly to the low-income areas.

Mohl and Hagen (2010) provides some evidence for the diversion of economic activity from the regions which were neighbouring the “objective 1” regions that were receiving additional funds from the ERDF, however, other studies such as Fiaschi et al (2018) suggest the opposite. While the displacement of businesses is certainly a valid concern, the conclusions form Gaubert (2018) are not generalisable to all geographically differentiated policies because they rely on a set of not fully realistic assumptions such as the perfect mobility of labour and the equalisation of utilities between all workers. They also do not take into account the dynamic long-term effects of the policies. Nonetheless, the model provides an interesting perspective and could serve as the basis to future theories which should also take into account the heterogeneity of firms.



4. Existing empirical studies

Even though the topic is relatively well-researched, various questions remain debated.

Besides, new data releases offer an opportunity to extend existing studies or apply methods that were not possible earlier. Researchers use varied model specifications and research setups, the most prevalent approach is regression analysis although some studies use other methods like propensity score matching or a regression discontinuity design. Over time, the studies increasingly refined their methodologies to accommodate for new techniques and more extensive statistics. Nevertheless, a large majority of the articles use regression specifications that are based on the neoclassical model. Most of the evaluations find positive but limited effects of the funding on growth and the convergence between regions (Pieńkowski and Berkowitz, 2016).

4.1 Empirical evaluations of the cohesion funds

Puigcerver-Peñalver (2007) surveys data from 15 European countries from 1989 to 1999 and infers that the funding contributed to regional growth and convergence during the first half of the studied period, while they remain unclear during the second. Esposti and Bussoletti (2008) examine the effects of “objective 1” funding which was accorded to regions with below 75% of the average European GDP per capita on their economic growth as well as convergence processes. The analysis spans across more than 200 regions and 11 years, from 1989 to 2000. The regressions are based on the neoclassical growth model. They conclude that the impact of the regional funds was mostly positive when considering the whole European Union, however, in some countries such as Spain, Germany and Greece, the funding had a negative impact. These mixed results implicate that there is a significant heterogeneity in outcomes across the different geographical areas. The negative results in some regions might be explained by the possible misallocations of funds and crowding-out of other public or private investments depending on the implementation process.

Mohl and Hagen (2010) employ various panel data approaches to figure out the impact that the EU structural funds in the years 1995-2005 had on regional growth. They disaggregate funds between the different objectives with objective 1 being the regional cohesion allocated to the poorest regions and objective 2 and 3 the regional competitivity. The regressions apply multiple lags to pick up effects that are distributed over time and whether they are stable. They do not find any statistically significant effects for the three combined objectives, which would indicate that in total, the funds are inefficient. However, the objective 1 grants seem to have a positive effect on growth. This means that the cohesion element was more effective than the overall structural funding. Adding the spatial lags to the equation to account for the interactions between regions does not change the conclusion, but the coefficient is smaller.

Maynou et al (2016) use a dynamic panel regression to estimate the effects of the European Regional Funds on the NUTS 2 regions between 1990 and 2010. They find that the grants have a statistically significant and positive impact on the GDP growth per capita. Increasing the funding by 1% will lead to an increase in economic growth by 0,9%. However, there were no statistically significant effects of the funds on output convergence which was occurring naturally at a varying rate. This indicates that the policy was not fully successful in its


15 objectives, achieving only its growth component without significantly reinforcing cohesion between areas.

Although the funding is accorded to specific localities, its impacts are likely to spread outside, to the neighbouring areas. An array of studies considers the spatial externalities, using matrices of physical and technological distances between regions to identify the spillovers in form of a common pool of specialized workers, better business connectiveness and transport infrastructure. Fiaschi et al (2018) use a geographically distributed neoclassical growth model for the EU12 countries in the 1989-2006 period. They find strong spatial externalities for the objective 1 regions, in that being a neighbour of such a region also yields considerable benefits. The results display additionality and the estimated multiplier is 1.52.

The effects are concave and bring statistically significant improvements up to 3%

expenditure per regional GDP and become insignificant from 4% upwards.

The 75% GDP threshold to access “objective 1” (currently “convergence objective”) funds from the ERDF offers an opportunity to use a Regression Discontinuity Design (RDD) which in this case, evaluates a discontinuity in a regression of growth on initial GDP per capita between areas that receive the cohesion funds (below the 75% cut-off) and those above.

The main benefits of this approach are that the results are easy to interpret and provide the direction of the causality. However, its core weakness is the use of a binary treatment variable and not differentiating based on the number of grants received. Therefore this method only estimates the impact of participating in the objective 1 funds and not the intensity of treatment received. This can be partially mitigated by applying the fuzzy RDD approach which ties the participation with the probability of obtaining funding, because being eligible is not always enough. Becker et al (2018) use this approach to assess the effects of structural funds on regional GDP growth in three programming periods between 1989 and 2013. They find that localities that become included gain 2.1 p.p growth, this would be an indication of the efficiency of the policy. However, regions that lose this status tend to lose 1.7 p.p. growth which raises additional questions about the sustainability of the gains acquired due to the cohesion funding.

4.2 The factors impacting the efficiency of grants

A growing literature considers what makes the grants efficient. Rodriguez-Posse and Garzillazo (2015) examine the role of institutions and find that regions with a greater quality of government obtain higher payoffs from the cohesion policy investments made. They regress the average economic growth on a range of explanatory variables including the initial GDP per capita level, the cohesion expenditure, a measure of the quality of government and on a set of controls. They find that the cohesion policy funds have a positive effect on growth which is heterogeneous among regions. The quality of institutions seems to be the most important factor determining the efficiency of grants because competent and impartial governance leads to superior investment decisions (Rodriguez-Posse 2013). This result was stable for all areas, regardless of their initial development and expenditures.

Cerqua and Pellegrini (2017) examine the effects of the intensity of the funds, taking into account the totality of the Structural Funds from 1994 to 2006. The authors estimate that the funds have a concave impact on the economic growth of regions. This implies that after


16 a certain threshold, this effect becomes negative and adding further funding only undermines the progress which was already achieved, because of the diminishing returns to investments as well as possible misallocations of funds due to their overabundance.

Furthermore, some entities might be too preoccupied with applying for additional grants to use the received funding in a productive way. There is a possible efficiency gain if the subsidies were reduced in 11 regions which received too much of it, which would also allow to save EUR 5.1 billion in EU funds during the studied period. Conclusions such as these have profound implications on the interpretations of the econometric results. The coefficients which estimate, are likely only valid for a given level expenditure and would be altered if the intensity of the funding changed.

The most recent developments in the field go beyond the study of regions and evaluate the firm-level effects of the cohesion policy. Fattorini et al. (2020) match the regionalised data on cohesion policy grants with the addresses of 273,500 European firms to examine what effects they had on firm-level productivity. The effects vary depending on the purpose of the funding. The targeted expenditure for research and development has a significant effect on total factor productivity while the generic business support funds fail to improve it. This stands in line with the endogenous growth model of Romer (1990). Moreover, the results indicate that the funding has largely heterogeneous effects depending on its purpose. In the private sector, the funding allocated for innovation has had a positive effect for firm productivity and therefore it was more likely to result in regional growth.

4.3 The ecological aspect of the cohesion subsidies

While discussing the implications of the cohesion funds on growth, it is vital to remember that the policy has various other objectives which do not necessarily align with it. Arguelles and Benavides (2014) evaluate the share of environmental spending of the structural funds in the 2000-2006 period. Curiously, they find that the “Objective 1” regions in southern European countries such as Spain, Italy and Portugal allocated a much larger share of spending to environmental conservation and ecological regeneration projects than similar regions located in Austria, Germany and the United Kingdom. They argue that this, the somewhat puzzling finding can be explained by a tradition of environmental spending in the latter countries, which did not need to allocate additional European funding for these purposes. Conversely, countries which did not fund these actions earlier tend to use the European resources. When analysing the spending on the environment, it is vital to bear in mind that the ecological improvement does not directly translate into GDP growth, therefore the full welfare effects of the structural funds are likely to be understated.

4.4 The implications of the empirical assessments

This literature review provides many useful insights into the effectiveness of cohesion funds.

Overall, they have had positive but limited effects, in most cases generating additional economic activity and contribute to the convergence of poorer regions (Pieńkowski and Berkowitz, 2016). Their effectiveness strongly depends on the quality of government and appropriate institutions. An important issue is that the funds’ effect on GDP growth is concave and added funding only boosts the GDP per capita up to a certain level of spending.

This is most likely due to the diminishing rate of return for undertaking more projects. There


17 are also questions about the sustainability of these gains, as losing the support can more than offset the earlier gains. Furthermore, it is important to note that not all of the funding has a direct effect on economic growth and can instead be allocated to other purposes such as environmental protection.

Most of the studies reviewed considered the sample of the EU15 countries, before the 2004 enlargement. There is not enough evidence on the effects of funding on economic growth including the countries which were admitted after that date. Therefore, performing an evaluation including the new joiners will help with filling this gap in the empirical literature.

Moreover, it could also help establish whether the changes in the programmes and larger geographical etendue affected their performance.



5. Data

The most important data source for this study is the “Historic EU payments - regionalised and modelled3” database, which provides the estimated EU regional funds expenditures from 1990 to 2018 for NUTS2 regions. Since the dataset has only been released in 2020, the timeframe studied can be extended in comparison to the previous studies. Additionally, the expenditure variable in the dataset used was modelled in order to better approximate the time when the payments took place on the ground, rather than when they were officially accorded. The payments are also disaggregated by region and the ERDF payments through the European Territorial Cooperation programmes, which cover multiple NUTS 2 units are excluded from the data. This information on the grants and loans can be matched with data from the Eurostat regional statistics bank4 in order to obtain the information on the economic outcomes and the required control variables. Using these sources allows for a sample size which is sufficient to draw reliable econometric inferences about the impact of the cohesion policy investments on the economic growth within the bounds of the sub- national territorial units.

The Eurostat regional statistics provide detailed information about the output levels, fixed capital accumulation the total number of employees and population educational attainment levels for NUTS 2. What complicates the matter, the data is presented in the NUTS 2016 format which prevents them from being fully matched with the regional policy data which is in NUTS 2013. With this inconsistency, only 86% of the region codes could be matched.

Therefore, the dataset was adjusted using the Eurostat NUTS conversion tool, using the daytime population as the chosen auxiliary covariate for the transformation. This allowed to match the data in 97% of cases. Some countries lack the data on the GDP, this is why the study does not consider the United Kingdom and France is only examined in the 2015-2018 period.

5.1 Sample

The European Union had its most important enlargement in 2004 when ten countries with combined populations of 75 million people joined the regional bloc. Since the regions which joined had a lower level per of GDP capita than the existing members, this event had a significant impact on the cohesion policy mechanisms, particularly by shifting the eligibility to additional ERDF funding from the regions of EU15 to the new member-states (Becker et al 2018). Due to these large divergences in the cohesion policy from before 2004 and after it, it is highly unlikely the relationship between the cohesion payments and the regional economic growth remained the same before and after that date. This is why the timeframe studied in this thesis begins in 2004, after the most important enlargements. Arguably, this overlooks the Accession of Romania and Bulgaria in 2007 and of Croatia in 2013. However, these changes were much smaller than the extension of 2004. The period examined finishes in 2018 due to the lack of data on the cohesion expenditure after that date.

The “Historic EU payments - regionalised and modelled” dataset contains information on the regional policy expenditure from 1989 to 2018, in all of the EU 271 NUTS2 regions, including

3 https://cohesiondata.ec.europa.eu/Other/Historic-EU-payments-regionalised-and-modelled/tc55-7ysv

4“Regional statistics by NUTS classification” database, Eurostat: https://ec.europa.eu/eurostat/web/regions/data/database


19 the United Kingdom before it left in 2020. The regions are presented in the NUTS 2013 format and provide information on the annual EU payments in the panel data format with yearly intervals and classified by region and between the distinct funds, including the ERDF, the CF and the ESF. The payments were also modelled and adjusted by the authors of the dataset to better reflect the dates when the expenditure actually happened, rather than when it was accorded (BERGEN, 2017). The data shows that in the 2004-2018 period, all of the regions studied received some support from at least one of those three funds and that their total expenditure amounted to 547 billion Euro. The majority of the funding (55.5%) was allocated through the ERDF, while 21.5% went through the CF and 23% through the ESF.

In order to perform the regression analysis, the expenditure dataset was joined to the regional economic and demographic variables from the Eurostat Regional Statistics Database. After using the Eurostat NUTS conversion tool to adjust the nomenclature to the NUTS 2016 standard, 97% of them could be matched. The sample used is also constrained by the availability of some regional data, lowering the final sample to 205 NUTS2 regions.

Notably, the economic figures are missing for the entirety of the United Kingdom which constitutes 37 out of the 66 missing regions. Furthermore, the regional output data for French territorial units is truncated and only available for the 2015-2018 period. However, the 205 regions matched received a total of 497 billion euro or 90.8% of the total funding.

Overall, the sample contains the GDP per worker and the control variables between 2004 and 2018 and the modelled expenditure of the three studied cohesion policy funds from 2001 to 2018, which was normalised by dividing them by the working populations of the regions in order to account for the differences in size. In the specifications which use lags of the cohesion expenditure, the regressions take into account additionally the subsidies originating before the studied period, i.e. from 2000 to 2003 in order not to lose observations due to lags.

The summary statistics of the expenditure variables used can be found in the table below.

Since the model uses a log transformation, they are presented in both their original and transformed forms. The logarithm of zero does not have a value for real numbers, therefore, the transformation led to some missing data for the regions which did not receive funding in a given year. In such cases, the missing values were replaced by zeroes to prevent the attrition.

Table 2: Descriptive Statistics

Variable Mean Std. Dev. Min Max

ERDF 123248.39 204075 0 3057289.8

CF 41084.957 88198.303 0 1102826.9

ESF 67469.429 159688.7 0 2499650.8

Log ERDF 10.036 2.931 0 14.933

Log CF 3.779 5.331 0 13.913

Log ESF 9.845 2.506 0 14.732

Number of observations: 3918


20 After all the limitations, there remains a total of 2767 observations distributed across 205 regions in a period of 14 years which are used in all of the regressions. Due to the structural similarity between the specifications, the number of observations is stable across all of them

6. The empirical exercise

This section presents the main part of the thesis, which is the econometric analysis. The regression specification is based on the neoclassical model which was first formulated and implemented by Solow (1956) and then augmented by Mankiw, Romer and Weil (1992) to account for the accumulation of human capital. Before proceeding to the estimation results, it is essential to discuss the key theoretical assumptions which are underpinning the model as well as the particularities of the estimation approach which are necessary for the correct interpretation of the obtained results.

The regression aims to identify the effects of the EU structural funds spending on the short- run regional economic growth while adjusting to the steady state. A key assumption in this model is that every region will converge towards their proper steady state which is conditional on the values of their human capital and investment rate. As described in sections 3.1 and 3.2, the production function depends on the total factor productivity parameter 𝐴, labour 𝐿, physical capital 𝐾 and human capital 𝐻.

𝑌 = 𝐴𝐹(𝐾; 𝐿; 𝐻)

In this case, the adjustment process includes additionally the accumulation of human capital through education and the acquisition of skills. The structural funds, especially the European Social Funds can provide additional educational opportunities and professional formation which, in turn will increase the steady state income of the targeted regions. Dummy variables for years capture the heterogeneity between years while region fixed effects are applied to adjust for time-invariant unobserved heterogeneity between the areas.

6.1 Regression specification

The regression specification is based on the adjustment path equation in the augmented version of the Solow model with an addition of the cohesion expenditure variable, similarly to the specification used by Mohl and Hagen (2010). The general version has been developed by Mankiw, Romer and Weil (1992) for cross-sectional data and adapted to a panel data form by Islam (1995). The fundamental version of the regression equation including cohesion policy funds, which is used in this thesis is presented below:

ln(𝑦𝑖,𝑡) − ln(𝑦𝑖,𝑡−1) = 𝛼𝑖 + 𝛾𝑡+ 𝛽1ln(𝑦𝑖,𝑡−1) + 𝛽2ln(𝑖𝑛𝑣𝑖,𝑡) + 𝛽2ln(𝑒𝑑𝑢𝑖,𝑡) + 𝛽3ln(𝑛𝑖,𝑡 + 𝑔 + 𝛿) + 𝛽4ln(𝑒𝑥𝑝𝑖,𝑡) +𝑢𝑖,𝑡

Where the subscript t signifies a year from 2004 to 2018 and i denotes a region i=1, …, 205 at a time t. The dependent variable is ∆ln (𝑦𝑖,𝑡) = ln(𝑦𝑖,𝑡) − ln (𝑦𝑖,𝑡−1) which approximates the rate of economic growth in percentage points. 𝑦𝑖,𝑡 is the Gross Domestic Product per effective worker, which allows the regression specification to more accurately represent the neoclassical growth model than the often-used GDP per capita. In order to adjust for different price level between countries it is measured in the purchasing power standard . 𝛼𝑖


21 and 𝛾𝑡 are the region and year fixed effects. 𝑖𝑛𝑣𝑖,𝑡 is the investment rate defined as the fixed capital accumulation to GDP ratio and 𝑒𝑑𝑢𝑖,𝑡 is the share of higher-educated people as a percentage of the regions’ population, which is used as a proxy for the human capital. 𝑛𝑖,𝑡 is the yearly growth rate of employment and the 𝑔 + 𝛿 are the technology growth and depreciation variables, assumed to be equal to 0.05 after Mankiw, Romer and Weil (1992).

𝑒𝑥𝑝𝑖,𝑡 – the modelled structural funds payments in a given region i at time t per effective worker. This variable was normalised dividing it by the regional employment and not by the GDP, because the latter can be correlated with the error term (Mohl and Hagen, 2010). The logarithm does not exist if there was no payments in a given year, in such cases, the missing values for ln(𝑒𝑥𝑝𝑖,𝑡) were replaced by 0. The error term 𝑢𝑖𝑡 accounts for the idiosyncratic errors. The GDP per effective employment is used rather than GDP per capita in order to better fit the neoclassical growth model. The regressions in tables 2 and 3 use the lags of the expenditure variable in order to determine how its effects are distributed over time and whether they are persistent. The regression specification becomes then:

∆ln (𝑦𝑖,𝑡)= 𝛼𝑖+ 𝛾𝑡+ 𝛽1ln(𝑦𝑖,𝑡−1)+ 𝛽2ln(𝑖𝑛𝑣𝑖,𝑡)+ 𝛽2ln(𝑒𝑑𝑢𝑖,𝑡)+ 𝛽3ln(𝑛𝑖,𝑡+ 𝑔 + 𝛿)+ 𝛽4ln(𝑒𝑥𝑝𝑖,𝑡)+ 𝛽5ln(𝑒𝑥𝑝𝑖,𝑡−1)+ 𝛽6ln(𝑒𝑥𝑝𝑖,𝑡−2)+ 𝛽7ln(𝑒𝑥𝑝𝑖,𝑡−3)+ 𝑢𝑖,𝑡

The model allows to distinguish between the immediate effect of cohesion subsidies in one year on short-term adjustment growth and their impact on the steady state output level if the funding is introduced permanently. The first are represented by the coefficients while the latter can be identified by taking a non-linear combinations of these estimates for the different lags and dividing them by the coefficient of the GDP per employment variable.

Applying logarithmic transformation on both the dependent and the independent variables allows the linear model to reflect the multiplicative relations present in the neoclassical growth model., such as the influence of the total factor productivity parameter on output as well as the exponential elements in the Cobb-Douglass production function. It also helps with mitigating the effect of severe outliers on the estimated coefficients (Woolridge 2015, p. 171). Another aspect of this functional form is that the coefficients estimate the proportional impact of a percentage change in independent variables to the dependent variable rather than the effect of an unit increase (Woolridge 2015, p. 172). Of course there are limitations, the logarithmic form is only an adequate approximation for small percentage changes such as the economic growth which rarely exceeds 10%.

6.2 Estimation approach

The estimated model is dynamic due to the fact that the lagged component of the dependent variable, ln(𝑦𝑖,𝑡−1), is also one of the explanatory variables. The standard estimators fail to account for the endogeneity caused by the dynamics within specification.

The lagged explanatory variable is correlated with the error term of the dependent variable because they share the same idiosyncratic shocks (Nickell, 1981). This bias can be remedied by using the two-stage least squares (2SLS) estimation using similar to the Anderson-Hsiao (1982) estimator. Using this technique, the first linear regression is run to estimate the instrumented variable using the instruments and then, the fitted variables are used as a regressor instead of the actual variable in the main specification.


22 The Nickell bias is especially cumbersome because it persists even with large sample sizes. It is therefore necessary to eliminate it with appropriate estimation techniques. The most commonly used method is the system General Method of Moments (GMM) estimator proposed by Arellano and Bond (1991). Both methods rely on the instrumental variable approach. A new and more efficient solution is proposed by Allison (2017) in form of a maximum likelihood model, however, it is not suited for longer time periods because of the high number of distinct calculations needed to obtain the results.

The instrumental variables (IV) approach is often used in cases where one of the explanatory variables (X) present in the regression is correlated with the error term. Using a standard estimation approach can yield results that would be biased (the estimator would not asymptotically approach the true value of the parameter) or inefficient (the estimator would have a high variance which may lead to rejecting true hypotheses or not rejecting false null hypotheses). The IV approach consists of first predicting the endogenous variable X using a non-endogenous variable Z and the fitted values 𝑋̂ in the main regression, which do no longer have an endogenous component. This can be done through a two-stage least squares (2SLS) estimation which is composed of two ordinary least squares regressions (OLS).

However, other estimators such as the GMM can also use the IV approach, which allows to eliminate the bias caused by the endogenous variables.

A large swatch of the variation in the data is results from the unobservable differences between the regions which can differ widely for example in their quality of institutions or their natural resources endowments. These factors have an impact on the outcome variable, however it can be removed using the fixed effects estimator. Such estimator subtracts the time averages of the variables from themselves in order to get rid of the time-invariant heterogeneity before estimating the model. This method is equivalent to the use of the least square dummy variable estimation, which allows for different intercept for each group in the panel. The data are also strongly fluctuating by year so additionally, year dummies are introduced as time fixed effects in order to control for irregularities such as the business cycles. One downside of the fixed effects estimator is that it removes all the time-invariant terms, therefore they could not be used as control variables.

6.3 Estimating the long-term effects

The individual estimates for the cohesion policy funds variable can only provide information about their influence on the short-term rate of economic growth approximated by the difference of logarithms of output ∆ln (𝑦𝑖,𝑡) = ln(𝑦𝑖,𝑡) − ln (𝑦𝑖,𝑡−1). It is necessary to bring multiple coefficients for lags of the variable to estimate long-term impact of the policy. In order to calculate the long-term effects of the cohesion spending on the output per employee, it is necessary to begin with the regression specification from section 6.1.

∆ln (𝑦𝑖,𝑡) = 𝛼𝑖+ 𝛾𝑡+ 𝛽1ln(𝑦𝑖,𝑡−1) + 𝛽2ln(𝑖𝑛𝑣𝑖,𝑡) + 𝛽2ln(𝑒𝑑𝑢𝑖,𝑡) + 𝛽3ln(𝑛𝑖,𝑡+ 𝑔 + 𝛿) + 𝛽4ln(𝑒𝑥𝑝𝑖,𝑡) + 𝛽5ln(𝑒𝑥𝑝𝑖,𝑡−1) + 𝛽6ln(𝑒𝑥𝑝𝑖,𝑡−2) + 𝛽7ln(𝑒𝑥𝑝𝑖,𝑡−3) + 𝑢𝑖,𝑡 For a given region 𝑖 at the time 𝑡.



In addition, the efficiency measure could forego (more) funds flowing to a NUTS-3 level region with high-growth opportunities, solely because it is part of a NUTS-2 level region

Like in the previous model with domestic value added shares, the immediate impact of structural funding in the same year and the next year conjectures that the economic integration,

Het  doel  van  het  onderzoek  was  de  inventarisatie  en  waardering  van  eventuele    archeologische  resten  die  door  de  geplande  bouwwerken 

Die dorp het wel na die beeidiging van die Eerste Vryheidsoor- log ·n onstabiele ekonomiese tydjie beleef maar het binne enkele maande herstel.. Die herstel van

Standard levels of fine particulate matter are classified by using different machine learning models. This classification is performed on six years’ records of daily

When the American doctrine is applied to the case of InnovaThor v Generix, InnovaThor’s Swiss-claim could be considered a patented medical treatment method. This method can be

Hiermee kunnen ziekteprocessen in het brein worden bestudeerd maar ook cognitieve processen zoals het waar- nemen van objecten of de betekenis van woorden in een

As die Afrikaner hom skuldig maak aan mense-verering, dan kom dit seker die duidelikste uit. Hierdie neiging moet teegegaan word-nie deur 'n reaksionere bouding